We analyze sequential detection for diffusion type signals both in the fixed probability of error formulation and in the Bayes Formulation. We show that the optimal strategy in both cases is of the threshold type with explicitly computable thresholds. We provide efficient numerical schemes for computing approximations to the likelihood ratio and provide an implementation via a special purpose VLSI processor for real time processing in the scalar diffusion case. Finally, we describe the DELPHI expert system which is under development. Its purpose is to provide an integrated system level design tool for sequential real time detection and estimation.